Deep Learning Based Automated Wound Detection for Clinical Image Analysis

Accurate and objective wound assessment is essential for effective clinical management, yet manual evaluation remains subjective, time consuming, and dependent on clinician expertise. In this study, we investigate a deep learning-based approach for automated wound segmentation and measurement using RGB images. A dataset of 3,960 wound photographs, combining three publicly available sources with clinically acquired images, was uniformly annotated at the pixel level and used to train an Attention U-Net model for robust wound boundary delineation. To enable real world dimensional analysis, a calibration marker placed adjacent to the wound is detected and utilized to estimate pixel to centimeter scale following geometric normalization. The segmented wound region is subsequently analyzed to compute key quantitative metrics, including area, perimeter, width, and height. Experimental evaluation demonstrates that the proposed approach provides accurate and consistent wound measurements across diverse clinical imaging conditions. These results highlight the potential of learning based segmentation combined with image based geometric analysis to support more objective wound assessment in clinical practice.